A multi-server queueing system is under study. Two kinds of requests are processed in the system. One kind of requests is generated by regular clients in exponentially distributed intervals of time. A whole sequence of requests can be generated by one regular client. Another kind of requests is generated by ad hoc clients. Each such client generates only one request. Regular and ad hoc client arrivals are defined by two independent Markov arrival processes. Service times for both kinds of requests are exponentially distributed, with the rate independent of the kind of request. A regular client resides in the system during an exponentially distributed time and departs from the system. All requests generated by this client when there are idle servers will obtain full service even the client has departed from the system. Ad hoc clients are less valuable for the system than the regular ones. Therefore, a request by a regular client is admitted to the system if there is at least one idle server at the moment of arrival. Ad hoc clients are admitted to the system only if the number of busy servers is less than a certain threshold value. The behaviour of this system is described by the four-dimensional continuous-time Markov chain with the state-inhomogeneous transition rates. Analysis of this chain is implemented, including demonstration of the feasibility of the proposed algorithms for computation of its stationary distribution. The problem of optimal choice of the number of servers and the threshold defining the policy of ad hoc clients admission is numerically solved.